Analysis Airbnb

import libraries

import plotly.offline as pyo
# Set notebook mode to work in offline
pyo.init_notebook_mode()


import pandas as pd
import matplotlib.pyplot as plt

import seaborn as sns
import plotly.express as px
import geopandas as gpd
import pprint

import numpy as np
#plt.rcParams['figure.figsize'] = [20, 15]

Collect Data

AirBnb Data

As of Jan 1, 2022, we can find 11 AirBnb data sets on http://insideairbnb.com/get-the-data.html, which were scrapped one time per month between Dec 2020 and Nov 2021, with May 2021 skipped. The 11 datasets where downloaded and put in 11 folders.

Reference Data

Get two reference data:

• List of confirmed Covid19 cases in NSW
  • retrieved from https://data.nsw.gov.au/search/dataset/ds-nsw-ckan-aefcde60-3b0c-4bc0-9af1-6fe652944ec2/details?q=
• Google Trends data regarding term Covid queried by Australian users
  • retrieved from https://trends.google.com/trends/explore?date=2020-11-01%202022-01-10&geo=AU&q=covid

We choose data of confirmed cases in NSW instead of Australia based on the assumption that the Covid situation in NSW has stronger impact on NSW AirBnb business than the Covid situations outside of NSW.

We choose data of Google Trends queried by users in Australia instead of NSW based on the assumption that NSW AirBnb guests are usually from all over Australia instead of NSW local residents. Opinions on Covid of people all over Australia may influence their decision on whether to book a NSW AirBnb property or not.

The numbers in Google Trends data indicate how the users' "interest" in term Covid changed during the given period. Roughly speaking, the highest frequency of search happening at a timepoint within the period is normalized to 100, and the frequencies happening at rest of the timepoints are normalzied to numbers <= 100, by comparing the frequencies against the highest.

# As of Jan 1, 2022, we can find 11 AirBnb data sets on http://insideairbnb.com/get-the-data.html, 
# which were scrapped one time per month between Dec 2020 and Nov 2021, with May 2021 skipped. 
# The 11 datasets where downloaded and put in 11 folders. The foler names are shown below.
AIRBNB_FOLDERS = ['data_airbnb_syd/20201214/',
           'data_airbnb_syd/20210110/',
           'data_airbnb_syd/20210208/',
           'data_airbnb_syd/20210304/',
           'data_airbnb_syd/20210410/',
           'data_airbnb_syd/20210605/',
           'data_airbnb_syd/20210705/',
           'data_airbnb_syd/20210807/',
           'data_airbnb_syd/20210908/',
           'data_airbnb_syd/20211009/',
           'data_airbnb_syd/20211106/']
# The file of confirmed Covid19 cases in NSW was downladed, and stored at location below
NSW_COVID_CASE_FILE = './data_covid/nsw_health/confirmed_cases_table1_location.csv'
# The file of Google Trends of query term 'Covid' was downloaded, and stored at location below
GOOGLE_TRENDS_COVID_FILE = './data_covid/google_trends/multiTimeline.csv'

1. AirBnb Operating Property Number and Occupancy

In this use case we estimate the number of AirBnb properties in operation and their occupancy rate in the period between Dec, 2020 and Nov, 2021. The data were collected by 11 scrapings, one scraping executed in each month, with May 2021 being skipped. One scraping usually lasted for 1-2 day.

We treat the status of properties at the time they were scraped as a snapshot, and therefore as a sample of their status in usual days. We estimate the number of properties in operation by counting the listed properties at scraping time, and whether the properties were occupied/vacant by checking if they were available for booking at scraping time.

The second estimation is based on a simplistic assumption that "available" means "vacant", and "unavailable" means "occupied". In reality, a property can be unavailable for two reasons, namely it is booked/occupied by guests, or the host of the property blocks booking. We assume the second reason for unavailable is not common, and can be safely ignored in the analysis.

We observe changes of property number and occapancy rate from the dataset. In pursuing the causes of the changes, we use two reference data, New South Wales confirmed Covid case data and Google Trends data on query term Covid.

Load and preprocess listing and calendar data in AirBnb dataset.

• retrieve scrape date of each invovled property
• retrieve whether a property was available when it was scraped

import os
available_property_nubers = []
unavailable_property_numbers = []
for folder in FOLDERS:
    date_label = folder.split('/')[1]
    last_scrape_date = pd.read_csv(os.path.join(folder, 'listings_detailed.csv'), 
                                   header='infer', 
                                   usecols=['id', 'calendar_last_scraped'], 
                                   parse_dates=['calendar_last_scraped'])
    last_scrape_date = last_scrape_date.rename(columns={'id': 'listing_id',
                                     'calendar_last_scraped': 'date'})
    df_cal_temp = pd.read_csv(os.path.join(folder, 'calendar_detailed.csv'), 
                              header='infer',
                              usecols=['listing_id', 'available', 'date'],
                              parse_dates=['date']
                             )
    df_cal_on_scrape_day = pd.merge(df_cal_temp, last_scrape_date, how='inner', on=['listing_id', 'date'])
    
    available_property_nubers.append((date_label, (df_cal_on_scrape_day['available']=='t').sum()))
    unavailable_property_numbers.append((date_label, (df_cal_on_scrape_day['available']=='f').sum()))
    
df_property_availability = pd.DataFrame({'date': [d for d, _ in available_property_nubers],
                                         'available': [a for _, a in available_property_nubers],
                                         'unavailable': [u for _, u in unavailable_property_numbers]})

df_property_availability['date'] = pd.to_datetime(df_property_availability['date'])
df_property_availability['property_number'] = df_property_availability['available'] + df_property_availability['unavailable']
df_property_availability['unavailable_ratio'] = df_property_availability['unavailable'] / df_property_availability['property_number']
df_property_availability['available_ratio'] = df_property_availability['available'] / df_property_availability['property_number']

Load and preprocess reference data.

• count number of newly confirmed Covid19 in NSW in each day within the period relevant to the anslysis
• retrieve Google Trends statistics of term Covid queried by users from Australia.

df_nsw_cases = pd.read_csv(NSW_COVID_CASE_FILE, header='infer', parse_dates=['notification_date'])
df_nsw_cases_count = df_nsw_cases[['notification_date']] 
df_nsw_cases_count = df_nsw_cases_count.groupby('notification_date').size().reset_index(name='counts')
df_nsw_cases_count = df_nsw_cases_count.loc[(df_nsw_cases_count['notification_date'] >= '2020-11-01') & (df_nsw_cases_count['notification_date'] <= '2021-11-30')]

df_trends = pd.read_csv(GOOGLE_TRENDS_COVID_FILE, header=1, parse_dates=['Week'])
#df_trends['covid: (Australia)'] = df_trends['covid: (Australia)'].str.replace(r'<1', '0')
df_trends['covid: (Australia)'] = df_trends['covid: (Australia)'].apply(pd.to_numeric) # latest version of plotly has a bug, needing explicitly changing object to numeric
df_trends = df_trends.loc[(df_trends['Week'] >= '2020-11-1') & (df_trends['Week'] <= '2021-11-30')]

Visualize the following in diagrams:

• NSW AirBnb listing number
• NSW AirBnb vacancy rate
• NSW Covid case number
• Australian people's interest in searching "covid" on Google

import plotly.graph_objects as go
from plotly.subplots import make_subplots

# Create figure with secondary y-axis
fig = make_subplots(rows=2, 
                    cols=1, 
                    specs=[[{"secondary_y": True}],
                           [{"secondary_y": True}]
                          ],
                    shared_xaxes=True,
                    vertical_spacing=0.02,
                    #subplot_titles=("NSW AirBnB Listing, Vacancy V.S. Confirmed Case, Google trends <I>Covid<I>")
                   )

# Add traces
fig.add_trace(
    go.Bar(x=df_property_availability['date'], y=df_property_availability['property_number'], name="Listed properties"),
    secondary_y=False,
    row=1,
    col=1
)

fig.add_trace(
    go.Bar(x=df_property_availability['date'], y=df_property_availability['available'], name="Vacant (~available) properties of AirBnb in NSW"),
    secondary_y=False,
    row=1,
    col=1
)


fig.add_trace(
    go.Scatter(x=df_property_availability['date'], y=df_property_availability['available_ratio'], name="Vacancy (~available) rate of AirBnb in NSW"),
    secondary_y=True,
    row=1,
    col=1
)



# Set x-axis title
#fig.update_xaxes(title_text="Sampling Date", row=1, col=1)

# Set y-axes titles
fig.update_yaxes(title_text="Property Number (log)", type='log', secondary_y=False, row=1, col=1)
fig.update_yaxes(title_text="Vacancy (~Available) Rate", range=[0, 0.5], secondary_y=True, row=1, col=1)

# Add traces
fig.add_trace(
    go.Bar(x=df_nsw_cases_count['notification_date'], y=df_nsw_cases_count['counts'], name="Confirmed cases in NSW"),
    secondary_y=False,
    row=2,
    col=1
)


fig.add_trace(
    go.Scatter(x=df_trends['Week'], y=df_trends['covid: (Australia)'], name="Australian users' interest in <I>Covid</I> (Google Trends)"),
    secondary_y=True,
    row=2,
    col=1
)


# Set x-axis title
#fig.update_xaxes(title_text="Date", row=2, col=1)

# Set y-axes titles
fig.update_yaxes(title_text="Confirmed case (log)", type='log', secondary_y=False, row=2, col=1)
fig.update_yaxes(title_text="Australian users' interest in <I>Covid</I> on Google", secondary_y=True, row=2, col=1)


# Add figure title
fig.update_layout(
    title_text="NSW AirBnB Listing, Vacancy V.S. Confirmed Case, Google trends <I>Covid</I>",
    height=800
)

# Set x-axis title
#fig.update_xaxes(title_text="Date")

# Set y-axes titles
#fig.update_yaxes(title_text="NSW Google search term Covid", type='log', secondary_y=True)
#fig.update_yaxes(title_text="NSW confirmed case number (Logorithm)", type='log', secondary_y=False)

fig.show()

Preliminary analysis of the diagram above

An initial and preliminary analysis of the diagram suggests:

• Vacancy rate of NSW AirBnb roughly complied with Australian people's interest in searching "Covid" (which in turn roughly indicates Australian people's concern of Covid), with a short period of lag.
  • For example, people's concern of Covid reached a very high level in July 2021, and the peak of NSW AirBnb vacancy rate arrived in August 2021.
• Australian people's concern of Covid became higher when NSW confirmed case number started to rise. People's concern may last for a while, and then dropped even if confirmed case number kept rising. It probably shows that people don't keep high alertness for long time, regardless of the circumtances.
• The number of properties listed in NSW AirBnb decreased very slowly from November 2020 to the beginning of October 2021, then the number dropped drastically in the following month. About 10k properties disappeared from AirBnb market in the period of one month. Whilst we have not reach a complete explanation to the drastic drop, we suggest one possible cuase is the recent rise in realestate market.
  • A part of the properties were probably put into the realestate market for sale. According to article https://www.afr.com/property/residential/new-listings-soar-by-41pc-in-sydney-as-vendors-rush-to-sell-20211101-p5952p, there was drastic increase in the number of listings in Sydney property market during October 2021.
    • Home listing increased by 25.5% to 29,183 in Sydney during October 2021.
    • In the inner west, 698 new properties were added during the month – a 52.5 per cent lift on existing listings
    • the eastern suburbs recorded 681 new homes for sale.
    • In the Hills district, 725 homes were listed, adding 24 per cent to the old stock
    • the upper north shore attracted 792 new listings – a 30 per cent increase over the month.

2. Analysis of property number and vacancy rate by neighbourhoods

TBD

3. Analysis of price

TBD

4. Predict price by description and/or review

TBD

is it achivable?

Below are messy material to be used for uses cases

look into listing.csv and listing.csv.gz

• File listing.csv is summary information and metrics for listings in Sydney (good for visualisations). Fields:
  • listing_id
  • id
  • date
  • reviewer_id
  • reviewer_name
  • comments
• File listing.csv.gz is detailed information for listings in Sydney. Fields:
  • id:   integer   Airbnb's unique identifier for the listing
  • name:  string
  • host_id:  integer
  • host_name:  string
  • neighbourhood_group:  text  Calculated  The neighbourhood group as geocoded using the latitude and longitude against neighborhoods as defined by open or public digital shapefiles.
  • neighbourhood:  text  Calculated  The neighbourhood as geocoded using the latitude and longitude against neighborhoods as defined by open or public digital shapefiles.
  • latitude:  numeric  Uses the World Geodetic System (WGS84) projection for latitude and longitude.
  • longitude:   Uses the World Geodetic System (WGS84) projection for latitude and longitude.
  • room_type:  string
  • price:  currency  daily price in local currency. Note, $ sign may be used despite locale
  • minimum_nights:  integer  minimum number of night stay for the listing (calendar rules may be different)
  • number_of_reviews:  integer  The number of reviews the listing has
  • last_review:  date  Calculated  The date of the last/newest review
  • calculated_host_listings_count:   integer   Calculated  The number of listings the host has in the current scrape, in the city/region geography.
  • availability_365:  integer   Calculated   avaliability_x.   The availability of the listing x days in the future as determined by the calendar. Note a listing may be available because it has been booked by a guest or blocked by the host.
  • number_of_reviews_ltm:  integer  Calculated   The number of reviews the listing has (in the last 12 months)
  • license:  string
• One row per property; properties are unique in file.

df_listings_summary_20211106.head()

#df_listings_summary_20211106.describe(include='all')

df_listings_summary_20211106['id'].nunique()

20513

df_listings_summary_20211106.query("availability_365 == 365").nunique()

df_listings_summary_20211106.query("availability_365 == 364").nunique()

for k, v in df_bnb_listings_detailed.iloc[0].iteritems():
    print(k,'\n' ,v)

df_bnb_listings_detailed.info()

df_bnb_cal_detailed.query("listing_id==11156")

df_bnb_cal_detailed.iloc[0]

Have a look into number of properties and number of hosts

df_listing_neighbourhood = df_listings_summary_20211106.groupby('neighbourhood')['id', 'host_id'].nunique()
df_listing_neighbourhood.reset_index(inplace=True)
#df_listing_neighbourhood

#fig, ax = plt.subplots(figsize=(18, 8))
#plt.xticks(rotation=75)
#ax = sns.barplot(data=df_listing_neighbourhood,
#                 x='neighbourhood', y='id')

#fig = px.bar(df_listing_neighbourhood,
#             x='neighbourhood', 
#             y='id',
#             hover_data=['id', 'host_id'],
#             labels={'id': 'number of properties', 'host_id': 'number of hosts'},
#             title='Number of properties in neighbourhoods',
#             color='id',
#             height=800
#            )
#fig.show()

#syd_geojson = gpd.read_file('./data_airbnb_syd/20211106/neighbourhoods.geojson')
#syd_geojson

#syd_geojson_joined = syd_geojson.merge(df_listing_neighbourhood, on='neighbourhood')
#syd_geojson_joined

#fig, axes = plt.subplots(2, figsize=(40, 25))
#syd_geojson_joined.plot(column='id', 
#                        ax=axes[0], 
#                        legend=True, 
#                        legend_kwds={'label': 'Number of properties', 'orientation': 'vertical'})
#
#syd_geojson_joined.plot(column='host_id', 
#                        ax=axes[1], 
#                        legend=True, 
#                        legend_kwds={'label': 'Number of hosts', 'orientation': 'vertical'})

#syd_geojson_joined.boundary.plot()

#syd_geojson_joined.centroid.plot()

#syd_geojson_joined.explore(column='id', 
#                           popup=['neighbourhood', 'id', 'host_id'],
#                           popup_kwds={'aliases': [('neighbourhood'), ('property number'), ('host number')]},
#                           tooltip=['neighbourhood', 'id', 'host_id'],
#                           tooltip_kwds={'aliases': [('neighbourhood'), ('property number'), ('host number')]},
#                           legend=True, 
#                           legend_kwds={'caption': 'Number of properties'})

#syd_geojson_joined.explore(column='host_id', 
#                           popup=['neighbourhood', 'id', 'host_id'],
#                           popup_kwds={'aliases': [('neighbourhood'), ('property number'), ('host number')]},
#                           tooltip=['neighbourhood', 'id', 'host_id'],
#                           tooltip_kwds={'aliases': [('neighbourhood'), ('property number'), ('host number')]},
#                           legend=True, 
#                           legend_kwds={'caption': 'Number of hosts'},
#                           k=2)

Get location (point) by name.

gpd.tools.geocode(syd_geojson['neighbourhood'])

look at listing prices

fig = px.violin(df_listings_summary_20211106, 
                y="price", 
                x="neighbourhood", 
                color="neighbourhood", 
                box=True, 
                points='outliers',#points="all",
                hover_data=['neighbourhood', 'price'],
                #width=2000,
                #height=2000,
                title='Price in neighbourhoods'
               )
fig.show()

df_listing_20211106_price = df_listings_summary_20211106.groupby('neighbourhood')['price'].agg([np.min, np.max, np.mean, np.median])
df_listing_20211106_price = df_listing_20211106_price.reset_index()

df_listing_20211106_price

df_listing_20211106_price.iloc[0]

#fig = px.choropleth(df_listing_20211106_price,
#                    geojson=syd_geojson,
#                    locations='neighbourhood',
#                    featureidkey='properties.neighbourhood',
#                    color='median',
#                    color_continuous_scale='Viridis',
#                    hover_name='neighbourhood',
#                    hover_data=['median', 'mean']
#                   )
#fig.update_layout(margin={"r":0, "t":0, "l": 0, "b": 0})
#fig.update_geos(fitbounds="locations")
#fig.show()

#dir(syd_geojson.centroid[0])

print(syd_geojson.centroid[0].coords[0])

## px.choropleth_mapbox does not calculate center and zoom automatically.
## randomly select one point within relevant area as the center of map.
#center_coords = syd_geojson.centroid[0].coords[0] # note: it is in format (point.x, point.y), so (longtitude, latitude)
#
#fig = px.choropleth_mapbox(df_listing_20211106_price,
#                    geojson=syd_geojson,
#                    locations='neighbourhood',
#                    featureidkey='properties.neighbourhood',
#                    color='median',
#                    color_continuous_scale='Viridis',
#                    hover_name='neighbourhood',
#                    hover_data=['median', 'mean'],
#                    mapbox_style="open-street-map",
#                    center={'lat': center_coords[1], 'lon': center_coords[0]} 
#                   )
#fig.update_layout(margin={"r":0, "t":0, "l": 0, "b": 0})
#fig.update_geos(fitbounds="locations", visible=False) ## it is not working on px.choropleth_mapbox. need to set center manually
#
#fig.show()

#center_coords = syd_geojson.centroid[0].coords[0]
#
#fig = px.choropleth_mapbox(df_listing_20211106_price, 
#                           geojson=syd_geojson,
#                           locations='neighbourhood',
#                           featureidkey='properties.neighbourhood',
#                           color='amax',
#                           color_continuous_scale="Viridis",   
#                           hover_name='neighbourhood',
#                           hover_data=['amax', 'amin'],
#                           center={'lat': center_coords[1], 'lon': center_coords[0]} 
#                          )
#fig.update_layout(margin={"r":0,"t":0,"l":0,"b":0})
#fig.update_layout(mapbox_style="open-street-map")  
#fig.update_geos(fitbounds="locations", visible=False)## it is not working on px.choropleth_mapbox. need to set center manually


# Note: if you put fig.update_layout() below instead of above (before fig.update_geos), 
# no data will be shown on map.
#fig.update_layout(mapbox_style="open-street-map")  

#fig.show()

import os
df_list = []
for folder in FOLDERS:
    date_label = folder.split('/')[1]
    file_path = os.path.join(folder, 'listings.csv') 
    df_temp = pd.read_csv(file_path, header='infer')
    df_temp = df_temp.groupby('neighbourhood')['price'].agg([np.min, np.max, np.mean, np.median])
    df_temp = df_temp.reset_index()
    df_temp = df_temp.assign(listing_date=date_label) 
    df_list.append(df_temp)

    
df_pricing = pd.concat(df_list)

df_pricing

df_pricing['listing_date_form'] = pd.to_datetime(df_pricing['listing_date'])
df_pricing['listing_date_form'] = df_pricing['listing_date_form'].astype(str)
df_pricing.sort_values('listing_date_form')
df_pricing

df_pricing.info()

fig = px.bar(df_pricing, 
             x="neighbourhood", 
             y="median", 
             color="neighbourhood",
             animation_frame="listing_date_form", 
             animation_group='neighbourhood', 
             range_x=[0,37],
             range_y=[0,600])

fig["layout"].pop("updatemenus") # optional, drop animation buttons
fig.show()

fig = px.bar(df_pricing, 
             x="neighbourhood", 
             y="amin", 
             color="neighbourhood",
             animation_frame="listing_date_form", 
             animation_group='neighbourhood', 
             range_x=[0,37],
             range_y=[0,100])

fig["layout"].pop("updatemenus") # optional, drop animation buttons
fig.show()

fig = px.bar(df_pricing, 
             x="neighbourhood", 
             y="amax", 
             color="neighbourhood",
             animation_frame="listing_date_form", 
             animation_group='neighbourhood', 
             range_x=[0,37],
             range_y=[1,100000],
             log_y=True
            )

fig["layout"].pop("updatemenus") # optional, drop animation buttons
fig.show()

df_listings_summary_20201214 = pd.read_csv("./data_airbnb_syd/20201214/listings.csv", header='infer')
df_listings_summary_20201214['id'].unique().shape

print('number of properties {}'.format(df_listings_summary_20201214['id'].unique().shape[0]))
print('number of hosts {}'.format(df_listings_summary_20201214['host_id'].unique().shape[0]))

import os
property_numbers = []
host_numbers = []
for folder in FOLDERS:
    date_label = folder.split('/')[1]
    file_path = os.path.join(folder, 'listings.csv') 
    df_temp = pd.read_csv(file_path, header='infer')
    property_numbers.append((date_label, df_temp['id'].unique().shape[0]))
    host_numbers.append((date_label, df_temp['host_id'].unique().shape[0]))

property_numbers

host_numbers

df_property_host_numbers = pd.DataFrame({'date': [d for d, _ in property_numbers],
                                         'property_number': [n for _, n in property_numbers],
                                         'host_number': [n for _, n in host_numbers]})

df_property_host_numbers

	date	property_number	host_number
0	20201214	33871	25321
1	20210110	33902	25097
2	20210208	33630	24965
3	20210304	33229	24759
4	20210410	33229	24759
5	20210605	32079	24055
6	20210705	31899	23913
7	20210807	31764	23799
8	20210908	31030	23467
9	20211009	30223	22980
10	20211106	20513	14977

#import matplotlib.pyplot as plt

#plt.xticks(rotation=45)
#plt.title('Property and Host numbers 2020 Dec - 2021 Nov')
#
#plt.plot(df_property_host_numbers['date'], df_property_host_numbers['property_number'], label='Property Number')
#plt.plot(df_property_host_numbers['date'], df_property_host_numbers['host_number'], label='Host Number')
#
#plt.legend()

#sns.relplot(data=df_property_host_numbers, kind='line')

#pd.melt(df_property_host_numbers, 
#        id_vars=['date'], 
#        value_vars=['property_number', 'host_number'],
#        var_name='number',
#       )

#sns.relplot(data=pd.melt(df_property_host_numbers, 
#                         id_vars=['date'], 
#                         value_vars=['property_number', 'host_number'],
#                         var_name='type',
#                         value_name='number'), 
#            x='date', 
#            y='number', 
#            hue='type', 
#            kind='line')

fig, ax = plt.subplots()
plt.title('Property and Host numbers 2020 Dec - 2021 Nov')
plt.xticks(rotation=45)
ax = sns.lineplot(data=df_property_host_numbers, x='date', y='property_number', label='Property Number')
ax1 = sns.lineplot(data=df_property_host_numbers, x='date', y='host_number', label='Host Number')
plt.xlabel('date')
plt.ylabel('number')
plt.legend()

df_listings_detailed_20211106 = pd.read_csv("./data_airbnb_syd/20211106/listings_detailed.csv", header='infer')

df_listings_detailed_20211106.head()

df_listings_detailed_20211106.info()

df_listings_detailed_20211106.iloc[0]

df_listings_detailed_20211106['calendar_last_scraped'].unique()

array(['2021-11-07', '2021-11-06'], dtype=object)

look into reviews.csv

• summary of reviews

look into neighbourhoods.csv and neighbourhoods.geojson

• neighbourhoods.csv is a list of suburb names within Great Sydney area
• neighbourhoods.geojson present the borders of each suburb in form of lists of geo coordinates

df_neighbourhoods_csv = pd.read_csv('./data_airbnb_syd/20210110/neighbourhoods.csv', header='infer')
df_neighbourhoods_csv.head()

df_neighbourhoods_csv.count()

df_neighbourhoods_csv

df_neighbourhoods_geo = pd.read_json('./data_airbnb_syd/20210110/neighbourhoods.geojson')
df_neighbourhoods_geo.head()

print(df_neighbourhoods_geo.iloc[0])
print(df_neighbourhoods_geo.iloc[0]['features'])